2 * linux/arch/x86_64/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz>
6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
18 #include <linux/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/initrd.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/proc_fs.h>
26 #include <linux/pci.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/memory.h>
31 #include <linux/memory_hotplug.h>
32 #include <linux/memremap.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
35 #include <linux/kcore.h>
37 #include <asm/processor.h>
38 #include <asm/bios_ebda.h>
39 #include <linux/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
43 #include <asm/fixmap.h>
44 #include <asm/e820/api.h>
47 #include <asm/mmu_context.h>
48 #include <asm/proto.h>
50 #include <asm/sections.h>
51 #include <asm/kdebug.h>
53 #include <asm/set_memory.h>
55 #include <asm/uv/uv.h>
56 #include <asm/setup.h>
58 #include "mm_internal.h"
60 #include "ident_map.c"
63 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
64 * physical space so we can cache the place of the first one and move
65 * around without checking the pgd every time.
68 /* Bits supported by the hardware: */
69 pteval_t __supported_pte_mask __read_mostly = ~0;
70 /* Bits allowed in normal kernel mappings: */
71 pteval_t __default_kernel_pte_mask __read_mostly = ~0;
72 EXPORT_SYMBOL_GPL(__supported_pte_mask);
73 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
74 EXPORT_SYMBOL(__default_kernel_pte_mask);
76 int force_personality32;
80 * Control non executable heap for 32bit processes.
81 * To control the stack too use noexec=off
83 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
84 * off PROT_READ implies PROT_EXEC
86 static int __init nonx32_setup(char *str)
88 if (!strcmp(str, "on"))
89 force_personality32 &= ~READ_IMPLIES_EXEC;
90 else if (!strcmp(str, "off"))
91 force_personality32 |= READ_IMPLIES_EXEC;
94 __setup("noexec32=", nonx32_setup);
96 static void sync_global_pgds_l5(unsigned long start, unsigned long end)
100 for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
101 const pgd_t *pgd_ref = pgd_offset_k(addr);
104 /* Check for overflow */
108 if (pgd_none(*pgd_ref))
111 spin_lock(&pgd_lock);
112 list_for_each_entry(page, &pgd_list, lru) {
114 spinlock_t *pgt_lock;
116 pgd = (pgd_t *)page_address(page) + pgd_index(addr);
117 /* the pgt_lock only for Xen */
118 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
121 if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
122 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
125 set_pgd(pgd, *pgd_ref);
127 spin_unlock(pgt_lock);
129 spin_unlock(&pgd_lock);
133 static void sync_global_pgds_l4(unsigned long start, unsigned long end)
137 for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
138 pgd_t *pgd_ref = pgd_offset_k(addr);
139 const p4d_t *p4d_ref;
143 * With folded p4d, pgd_none() is always false, we need to
144 * handle synchonization on p4d level.
146 MAYBE_BUILD_BUG_ON(pgd_none(*pgd_ref));
147 p4d_ref = p4d_offset(pgd_ref, addr);
149 if (p4d_none(*p4d_ref))
152 spin_lock(&pgd_lock);
153 list_for_each_entry(page, &pgd_list, lru) {
156 spinlock_t *pgt_lock;
158 pgd = (pgd_t *)page_address(page) + pgd_index(addr);
159 p4d = p4d_offset(pgd, addr);
160 /* the pgt_lock only for Xen */
161 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
164 if (!p4d_none(*p4d_ref) && !p4d_none(*p4d))
165 BUG_ON(p4d_page_vaddr(*p4d)
166 != p4d_page_vaddr(*p4d_ref));
169 set_p4d(p4d, *p4d_ref);
171 spin_unlock(pgt_lock);
173 spin_unlock(&pgd_lock);
178 * When memory was added make sure all the processes MM have
179 * suitable PGD entries in the local PGD level page.
181 void sync_global_pgds(unsigned long start, unsigned long end)
183 if (pgtable_l5_enabled())
184 sync_global_pgds_l5(start, end);
186 sync_global_pgds_l4(start, end);
190 * NOTE: This function is marked __ref because it calls __init function
191 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
193 static __ref void *spp_getpage(void)
198 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
200 ptr = alloc_bootmem_pages(PAGE_SIZE);
202 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
203 panic("set_pte_phys: cannot allocate page data %s\n",
204 after_bootmem ? "after bootmem" : "");
207 pr_debug("spp_getpage %p\n", ptr);
212 static p4d_t *fill_p4d(pgd_t *pgd, unsigned long vaddr)
214 if (pgd_none(*pgd)) {
215 p4d_t *p4d = (p4d_t *)spp_getpage();
216 pgd_populate(&init_mm, pgd, p4d);
217 if (p4d != p4d_offset(pgd, 0))
218 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
219 p4d, p4d_offset(pgd, 0));
221 return p4d_offset(pgd, vaddr);
224 static pud_t *fill_pud(p4d_t *p4d, unsigned long vaddr)
226 if (p4d_none(*p4d)) {
227 pud_t *pud = (pud_t *)spp_getpage();
228 p4d_populate(&init_mm, p4d, pud);
229 if (pud != pud_offset(p4d, 0))
230 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
231 pud, pud_offset(p4d, 0));
233 return pud_offset(p4d, vaddr);
236 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
238 if (pud_none(*pud)) {
239 pmd_t *pmd = (pmd_t *) spp_getpage();
240 pud_populate(&init_mm, pud, pmd);
241 if (pmd != pmd_offset(pud, 0))
242 printk(KERN_ERR "PAGETABLE BUG #02! %p <-> %p\n",
243 pmd, pmd_offset(pud, 0));
245 return pmd_offset(pud, vaddr);
248 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
250 if (pmd_none(*pmd)) {
251 pte_t *pte = (pte_t *) spp_getpage();
252 pmd_populate_kernel(&init_mm, pmd, pte);
253 if (pte != pte_offset_kernel(pmd, 0))
254 printk(KERN_ERR "PAGETABLE BUG #03!\n");
256 return pte_offset_kernel(pmd, vaddr);
259 static void __set_pte_vaddr(pud_t *pud, unsigned long vaddr, pte_t new_pte)
261 pmd_t *pmd = fill_pmd(pud, vaddr);
262 pte_t *pte = fill_pte(pmd, vaddr);
264 set_pte(pte, new_pte);
267 * It's enough to flush this one mapping.
268 * (PGE mappings get flushed as well)
270 __flush_tlb_one_kernel(vaddr);
273 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte)
275 p4d_t *p4d = p4d_page + p4d_index(vaddr);
276 pud_t *pud = fill_pud(p4d, vaddr);
278 __set_pte_vaddr(pud, vaddr, new_pte);
281 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
283 pud_t *pud = pud_page + pud_index(vaddr);
285 __set_pte_vaddr(pud, vaddr, new_pte);
288 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
293 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
295 pgd = pgd_offset_k(vaddr);
296 if (pgd_none(*pgd)) {
298 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
302 p4d_page = p4d_offset(pgd, 0);
303 set_pte_vaddr_p4d(p4d_page, vaddr, pteval);
306 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
312 pgd = pgd_offset_k(vaddr);
313 p4d = fill_p4d(pgd, vaddr);
314 pud = fill_pud(p4d, vaddr);
315 return fill_pmd(pud, vaddr);
318 pte_t * __init populate_extra_pte(unsigned long vaddr)
322 pmd = populate_extra_pmd(vaddr);
323 return fill_pte(pmd, vaddr);
327 * Create large page table mappings for a range of physical addresses.
329 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
330 enum page_cache_mode cache)
338 pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
339 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
340 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
341 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
342 pgd = pgd_offset_k((unsigned long)__va(phys));
343 if (pgd_none(*pgd)) {
344 p4d = (p4d_t *) spp_getpage();
345 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE |
348 p4d = p4d_offset(pgd, (unsigned long)__va(phys));
349 if (p4d_none(*p4d)) {
350 pud = (pud_t *) spp_getpage();
351 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE |
354 pud = pud_offset(p4d, (unsigned long)__va(phys));
355 if (pud_none(*pud)) {
356 pmd = (pmd_t *) spp_getpage();
357 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
360 pmd = pmd_offset(pud, phys);
361 BUG_ON(!pmd_none(*pmd));
362 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
366 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
368 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
371 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
373 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
377 * The head.S code sets up the kernel high mapping:
379 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
381 * phys_base holds the negative offset to the kernel, which is added
382 * to the compile time generated pmds. This results in invalid pmds up
383 * to the point where we hit the physaddr 0 mapping.
385 * We limit the mappings to the region from _text to _brk_end. _brk_end
386 * is rounded up to the 2MB boundary. This catches the invalid pmds as
387 * well, as they are located before _text:
389 void __init cleanup_highmap(void)
391 unsigned long vaddr = __START_KERNEL_map;
392 unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
393 unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
394 pmd_t *pmd = level2_kernel_pgt;
397 * Native path, max_pfn_mapped is not set yet.
398 * Xen has valid max_pfn_mapped set in
399 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
402 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
404 for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
407 if (vaddr < (unsigned long) _text || vaddr > end)
408 set_pmd(pmd, __pmd(0));
413 * Create PTE level page table mapping for physical addresses.
414 * It returns the last physical address mapped.
416 static unsigned long __meminit
417 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
420 unsigned long pages = 0, paddr_next;
421 unsigned long paddr_last = paddr_end;
425 pte = pte_page + pte_index(paddr);
426 i = pte_index(paddr);
428 for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
429 paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
430 if (paddr >= paddr_end) {
431 if (!after_bootmem &&
432 !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
434 !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
435 E820_TYPE_RESERVED_KERN))
436 set_pte(pte, __pte(0));
441 * We will re-use the existing mapping.
442 * Xen for example has some special requirements, like mapping
443 * pagetable pages as RO. So assume someone who pre-setup
444 * these mappings are more intelligent.
446 if (!pte_none(*pte)) {
453 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte, paddr,
454 pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
456 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
457 paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
460 update_page_count(PG_LEVEL_4K, pages);
466 * Create PMD level page table mapping for physical addresses. The virtual
467 * and physical address have to be aligned at this level.
468 * It returns the last physical address mapped.
470 static unsigned long __meminit
471 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
472 unsigned long page_size_mask, pgprot_t prot)
474 unsigned long pages = 0, paddr_next;
475 unsigned long paddr_last = paddr_end;
477 int i = pmd_index(paddr);
479 for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
480 pmd_t *pmd = pmd_page + pmd_index(paddr);
482 pgprot_t new_prot = prot;
484 paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
485 if (paddr >= paddr_end) {
486 if (!after_bootmem &&
487 !e820__mapped_any(paddr & PMD_MASK, paddr_next,
489 !e820__mapped_any(paddr & PMD_MASK, paddr_next,
490 E820_TYPE_RESERVED_KERN))
491 set_pmd(pmd, __pmd(0));
495 if (!pmd_none(*pmd)) {
496 if (!pmd_large(*pmd)) {
497 spin_lock(&init_mm.page_table_lock);
498 pte = (pte_t *)pmd_page_vaddr(*pmd);
499 paddr_last = phys_pte_init(pte, paddr,
501 spin_unlock(&init_mm.page_table_lock);
505 * If we are ok with PG_LEVEL_2M mapping, then we will
506 * use the existing mapping,
508 * Otherwise, we will split the large page mapping but
509 * use the same existing protection bits except for
510 * large page, so that we don't violate Intel's TLB
511 * Application note (317080) which says, while changing
512 * the page sizes, new and old translations should
513 * not differ with respect to page frame and
516 if (page_size_mask & (1 << PG_LEVEL_2M)) {
519 paddr_last = paddr_next;
522 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
525 if (page_size_mask & (1<<PG_LEVEL_2M)) {
527 spin_lock(&init_mm.page_table_lock);
528 set_pte((pte_t *)pmd,
529 pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
530 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
531 spin_unlock(&init_mm.page_table_lock);
532 paddr_last = paddr_next;
536 pte = alloc_low_page();
537 paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot);
539 spin_lock(&init_mm.page_table_lock);
540 pmd_populate_kernel(&init_mm, pmd, pte);
541 spin_unlock(&init_mm.page_table_lock);
543 update_page_count(PG_LEVEL_2M, pages);
548 * Create PUD level page table mapping for physical addresses. The virtual
549 * and physical address do not have to be aligned at this level. KASLR can
550 * randomize virtual addresses up to this level.
551 * It returns the last physical address mapped.
553 static unsigned long __meminit
554 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
555 unsigned long page_size_mask)
557 unsigned long pages = 0, paddr_next;
558 unsigned long paddr_last = paddr_end;
559 unsigned long vaddr = (unsigned long)__va(paddr);
560 int i = pud_index(vaddr);
562 for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
565 pgprot_t prot = PAGE_KERNEL;
567 vaddr = (unsigned long)__va(paddr);
568 pud = pud_page + pud_index(vaddr);
569 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
571 if (paddr >= paddr_end) {
572 if (!after_bootmem &&
573 !e820__mapped_any(paddr & PUD_MASK, paddr_next,
575 !e820__mapped_any(paddr & PUD_MASK, paddr_next,
576 E820_TYPE_RESERVED_KERN))
577 set_pud(pud, __pud(0));
581 if (!pud_none(*pud)) {
582 if (!pud_large(*pud)) {
583 pmd = pmd_offset(pud, 0);
584 paddr_last = phys_pmd_init(pmd, paddr,
592 * If we are ok with PG_LEVEL_1G mapping, then we will
593 * use the existing mapping.
595 * Otherwise, we will split the gbpage mapping but use
596 * the same existing protection bits except for large
597 * page, so that we don't violate Intel's TLB
598 * Application note (317080) which says, while changing
599 * the page sizes, new and old translations should
600 * not differ with respect to page frame and
603 if (page_size_mask & (1 << PG_LEVEL_1G)) {
606 paddr_last = paddr_next;
609 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
612 if (page_size_mask & (1<<PG_LEVEL_1G)) {
614 spin_lock(&init_mm.page_table_lock);
615 set_pte((pte_t *)pud,
616 pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
618 spin_unlock(&init_mm.page_table_lock);
619 paddr_last = paddr_next;
623 pmd = alloc_low_page();
624 paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
625 page_size_mask, prot);
627 spin_lock(&init_mm.page_table_lock);
628 pud_populate(&init_mm, pud, pmd);
629 spin_unlock(&init_mm.page_table_lock);
633 update_page_count(PG_LEVEL_1G, pages);
638 static unsigned long __meminit
639 phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
640 unsigned long page_size_mask)
642 unsigned long paddr_next, paddr_last = paddr_end;
643 unsigned long vaddr = (unsigned long)__va(paddr);
644 int i = p4d_index(vaddr);
646 if (!pgtable_l5_enabled())
647 return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end, page_size_mask);
649 for (; i < PTRS_PER_P4D; i++, paddr = paddr_next) {
653 vaddr = (unsigned long)__va(paddr);
654 p4d = p4d_page + p4d_index(vaddr);
655 paddr_next = (paddr & P4D_MASK) + P4D_SIZE;
657 if (paddr >= paddr_end) {
658 if (!after_bootmem &&
659 !e820__mapped_any(paddr & P4D_MASK, paddr_next,
661 !e820__mapped_any(paddr & P4D_MASK, paddr_next,
662 E820_TYPE_RESERVED_KERN))
663 set_p4d(p4d, __p4d(0));
667 if (!p4d_none(*p4d)) {
668 pud = pud_offset(p4d, 0);
669 paddr_last = phys_pud_init(pud, paddr,
676 pud = alloc_low_page();
677 paddr_last = phys_pud_init(pud, paddr, paddr_end,
680 spin_lock(&init_mm.page_table_lock);
681 p4d_populate(&init_mm, p4d, pud);
682 spin_unlock(&init_mm.page_table_lock);
690 * Create page table mapping for the physical memory for specific physical
691 * addresses. The virtual and physical addresses have to be aligned on PMD level
692 * down. It returns the last physical address mapped.
694 unsigned long __meminit
695 kernel_physical_mapping_init(unsigned long paddr_start,
696 unsigned long paddr_end,
697 unsigned long page_size_mask)
699 bool pgd_changed = false;
700 unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
702 paddr_last = paddr_end;
703 vaddr = (unsigned long)__va(paddr_start);
704 vaddr_end = (unsigned long)__va(paddr_end);
707 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
708 pgd_t *pgd = pgd_offset_k(vaddr);
711 vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
714 p4d = (p4d_t *)pgd_page_vaddr(*pgd);
715 paddr_last = phys_p4d_init(p4d, __pa(vaddr),
721 p4d = alloc_low_page();
722 paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
725 spin_lock(&init_mm.page_table_lock);
726 if (pgtable_l5_enabled())
727 pgd_populate(&init_mm, pgd, p4d);
729 p4d_populate(&init_mm, p4d_offset(pgd, vaddr), (pud_t *) p4d);
730 spin_unlock(&init_mm.page_table_lock);
735 sync_global_pgds(vaddr_start, vaddr_end - 1);
743 void __init initmem_init(void)
745 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
749 void __init paging_init(void)
751 sparse_memory_present_with_active_regions(MAX_NUMNODES);
755 * clear the default setting with node 0
756 * note: don't use nodes_clear here, that is really clearing when
757 * numa support is not compiled in, and later node_set_state
758 * will not set it back.
760 node_clear_state(0, N_MEMORY);
761 if (N_MEMORY != N_NORMAL_MEMORY)
762 node_clear_state(0, N_NORMAL_MEMORY);
768 * Memory hotplug specific functions
770 #ifdef CONFIG_MEMORY_HOTPLUG
772 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
775 static void update_end_of_memory_vars(u64 start, u64 size)
777 unsigned long end_pfn = PFN_UP(start + size);
779 if (end_pfn > max_pfn) {
781 max_low_pfn = end_pfn;
782 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
786 int add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
787 struct vmem_altmap *altmap, bool want_memblock)
791 ret = __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
794 /* update max_pfn, max_low_pfn and high_memory */
795 update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
796 nr_pages << PAGE_SHIFT);
801 int arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
804 unsigned long start_pfn = start >> PAGE_SHIFT;
805 unsigned long nr_pages = size >> PAGE_SHIFT;
807 init_memory_mapping(start, start + size);
809 return add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
812 #define PAGE_INUSE 0xFD
814 static void __meminit free_pagetable(struct page *page, int order)
817 unsigned int nr_pages = 1 << order;
819 /* bootmem page has reserved flag */
820 if (PageReserved(page)) {
821 __ClearPageReserved(page);
823 magic = (unsigned long)page->freelist;
824 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
826 put_page_bootmem(page++);
829 free_reserved_page(page++);
831 free_pages((unsigned long)page_address(page), order);
834 static void __meminit free_hugepage_table(struct page *page,
835 struct vmem_altmap *altmap)
838 vmem_altmap_free(altmap, PMD_SIZE / PAGE_SIZE);
840 free_pagetable(page, get_order(PMD_SIZE));
843 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
848 for (i = 0; i < PTRS_PER_PTE; i++) {
854 /* free a pte talbe */
855 free_pagetable(pmd_page(*pmd), 0);
856 spin_lock(&init_mm.page_table_lock);
858 spin_unlock(&init_mm.page_table_lock);
861 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
866 for (i = 0; i < PTRS_PER_PMD; i++) {
872 /* free a pmd talbe */
873 free_pagetable(pud_page(*pud), 0);
874 spin_lock(&init_mm.page_table_lock);
876 spin_unlock(&init_mm.page_table_lock);
879 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
884 for (i = 0; i < PTRS_PER_PUD; i++) {
890 /* free a pud talbe */
891 free_pagetable(p4d_page(*p4d), 0);
892 spin_lock(&init_mm.page_table_lock);
894 spin_unlock(&init_mm.page_table_lock);
897 static void __meminit
898 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
901 unsigned long next, pages = 0;
904 phys_addr_t phys_addr;
906 pte = pte_start + pte_index(addr);
907 for (; addr < end; addr = next, pte++) {
908 next = (addr + PAGE_SIZE) & PAGE_MASK;
912 if (!pte_present(*pte))
916 * We mapped [0,1G) memory as identity mapping when
917 * initializing, in arch/x86/kernel/head_64.S. These
918 * pagetables cannot be removed.
920 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
921 if (phys_addr < (phys_addr_t)0x40000000)
924 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
926 * Do not free direct mapping pages since they were
927 * freed when offlining, or simplely not in use.
930 free_pagetable(pte_page(*pte), 0);
932 spin_lock(&init_mm.page_table_lock);
933 pte_clear(&init_mm, addr, pte);
934 spin_unlock(&init_mm.page_table_lock);
936 /* For non-direct mapping, pages means nothing. */
940 * If we are here, we are freeing vmemmap pages since
941 * direct mapped memory ranges to be freed are aligned.
943 * If we are not removing the whole page, it means
944 * other page structs in this page are being used and
945 * we canot remove them. So fill the unused page_structs
946 * with 0xFD, and remove the page when it is wholly
949 memset((void *)addr, PAGE_INUSE, next - addr);
951 page_addr = page_address(pte_page(*pte));
952 if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
953 free_pagetable(pte_page(*pte), 0);
955 spin_lock(&init_mm.page_table_lock);
956 pte_clear(&init_mm, addr, pte);
957 spin_unlock(&init_mm.page_table_lock);
962 /* Call free_pte_table() in remove_pmd_table(). */
965 update_page_count(PG_LEVEL_4K, -pages);
968 static void __meminit
969 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
970 bool direct, struct vmem_altmap *altmap)
972 unsigned long next, pages = 0;
977 pmd = pmd_start + pmd_index(addr);
978 for (; addr < end; addr = next, pmd++) {
979 next = pmd_addr_end(addr, end);
981 if (!pmd_present(*pmd))
984 if (pmd_large(*pmd)) {
985 if (IS_ALIGNED(addr, PMD_SIZE) &&
986 IS_ALIGNED(next, PMD_SIZE)) {
988 free_hugepage_table(pmd_page(*pmd),
991 spin_lock(&init_mm.page_table_lock);
993 spin_unlock(&init_mm.page_table_lock);
996 /* If here, we are freeing vmemmap pages. */
997 memset((void *)addr, PAGE_INUSE, next - addr);
999 page_addr = page_address(pmd_page(*pmd));
1000 if (!memchr_inv(page_addr, PAGE_INUSE,
1002 free_hugepage_table(pmd_page(*pmd),
1005 spin_lock(&init_mm.page_table_lock);
1007 spin_unlock(&init_mm.page_table_lock);
1014 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
1015 remove_pte_table(pte_base, addr, next, direct);
1016 free_pte_table(pte_base, pmd);
1019 /* Call free_pmd_table() in remove_pud_table(). */
1021 update_page_count(PG_LEVEL_2M, -pages);
1024 static void __meminit
1025 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
1026 struct vmem_altmap *altmap, bool direct)
1028 unsigned long next, pages = 0;
1033 pud = pud_start + pud_index(addr);
1034 for (; addr < end; addr = next, pud++) {
1035 next = pud_addr_end(addr, end);
1037 if (!pud_present(*pud))
1040 if (pud_large(*pud)) {
1041 if (IS_ALIGNED(addr, PUD_SIZE) &&
1042 IS_ALIGNED(next, PUD_SIZE)) {
1044 free_pagetable(pud_page(*pud),
1045 get_order(PUD_SIZE));
1047 spin_lock(&init_mm.page_table_lock);
1049 spin_unlock(&init_mm.page_table_lock);
1052 /* If here, we are freeing vmemmap pages. */
1053 memset((void *)addr, PAGE_INUSE, next - addr);
1055 page_addr = page_address(pud_page(*pud));
1056 if (!memchr_inv(page_addr, PAGE_INUSE,
1058 free_pagetable(pud_page(*pud),
1059 get_order(PUD_SIZE));
1061 spin_lock(&init_mm.page_table_lock);
1063 spin_unlock(&init_mm.page_table_lock);
1070 pmd_base = pmd_offset(pud, 0);
1071 remove_pmd_table(pmd_base, addr, next, direct, altmap);
1072 free_pmd_table(pmd_base, pud);
1076 update_page_count(PG_LEVEL_1G, -pages);
1079 static void __meminit
1080 remove_p4d_table(p4d_t *p4d_start, unsigned long addr, unsigned long end,
1081 struct vmem_altmap *altmap, bool direct)
1083 unsigned long next, pages = 0;
1087 p4d = p4d_start + p4d_index(addr);
1088 for (; addr < end; addr = next, p4d++) {
1089 next = p4d_addr_end(addr, end);
1091 if (!p4d_present(*p4d))
1094 BUILD_BUG_ON(p4d_large(*p4d));
1096 pud_base = pud_offset(p4d, 0);
1097 remove_pud_table(pud_base, addr, next, altmap, direct);
1099 * For 4-level page tables we do not want to free PUDs, but in the
1100 * 5-level case we should free them. This code will have to change
1101 * to adapt for boot-time switching between 4 and 5 level page tables.
1103 if (pgtable_l5_enabled())
1104 free_pud_table(pud_base, p4d);
1108 update_page_count(PG_LEVEL_512G, -pages);
1111 /* start and end are both virtual address. */
1112 static void __meminit
1113 remove_pagetable(unsigned long start, unsigned long end, bool direct,
1114 struct vmem_altmap *altmap)
1121 for (addr = start; addr < end; addr = next) {
1122 next = pgd_addr_end(addr, end);
1124 pgd = pgd_offset_k(addr);
1125 if (!pgd_present(*pgd))
1128 p4d = p4d_offset(pgd, 0);
1129 remove_p4d_table(p4d, addr, next, altmap, direct);
1135 void __ref vmemmap_free(unsigned long start, unsigned long end,
1136 struct vmem_altmap *altmap)
1138 remove_pagetable(start, end, false, altmap);
1141 #ifdef CONFIG_MEMORY_HOTREMOVE
1142 static void __meminit
1143 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1145 start = (unsigned long)__va(start);
1146 end = (unsigned long)__va(end);
1148 remove_pagetable(start, end, true, NULL);
1151 int __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1153 unsigned long start_pfn = start >> PAGE_SHIFT;
1154 unsigned long nr_pages = size >> PAGE_SHIFT;
1155 struct page *page = pfn_to_page(start_pfn);
1159 /* With altmap the first mapped page is offset from @start */
1161 page += vmem_altmap_offset(altmap);
1162 zone = page_zone(page);
1163 ret = __remove_pages(zone, start_pfn, nr_pages, altmap);
1165 kernel_physical_mapping_remove(start, start + size);
1170 #endif /* CONFIG_MEMORY_HOTPLUG */
1172 static struct kcore_list kcore_vsyscall;
1174 static void __init register_page_bootmem_info(void)
1179 for_each_online_node(i)
1180 register_page_bootmem_info_node(NODE_DATA(i));
1184 void __init mem_init(void)
1188 /* clear_bss() already clear the empty_zero_page */
1190 /* this will put all memory onto the freelists */
1193 x86_init.hyper.init_after_bootmem();
1196 * Must be done after boot memory is put on freelist, because here we
1197 * might set fields in deferred struct pages that have not yet been
1198 * initialized, and free_all_bootmem() initializes all the reserved
1199 * deferred pages for us.
1201 register_page_bootmem_info();
1203 /* Register memory areas for /proc/kcore */
1204 if (get_gate_vma(&init_mm))
1205 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR, PAGE_SIZE, KCORE_USER);
1207 mem_init_print_info(NULL);
1210 int kernel_set_to_readonly;
1212 void set_kernel_text_rw(void)
1214 unsigned long start = PFN_ALIGN(_text);
1215 unsigned long end = PFN_ALIGN(__stop___ex_table);
1217 if (!kernel_set_to_readonly)
1220 pr_debug("Set kernel text: %lx - %lx for read write\n",
1224 * Make the kernel identity mapping for text RW. Kernel text
1225 * mapping will always be RO. Refer to the comment in
1226 * static_protections() in pageattr.c
1228 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1231 void set_kernel_text_ro(void)
1233 unsigned long start = PFN_ALIGN(_text);
1234 unsigned long end = PFN_ALIGN(__stop___ex_table);
1236 if (!kernel_set_to_readonly)
1239 pr_debug("Set kernel text: %lx - %lx for read only\n",
1243 * Set the kernel identity mapping for text RO.
1245 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1248 void mark_rodata_ro(void)
1250 unsigned long start = PFN_ALIGN(_text);
1251 unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1252 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1253 unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1254 unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1255 unsigned long all_end;
1257 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1258 (end - start) >> 10);
1259 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1261 kernel_set_to_readonly = 1;
1264 * The rodata/data/bss/brk section (but not the kernel text!)
1265 * should also be not-executable.
1267 * We align all_end to PMD_SIZE because the existing mapping
1268 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1269 * split the PMD and the reminder between _brk_end and the end
1270 * of the PMD will remain mapped executable.
1272 * Any PMD which was setup after the one which covers _brk_end
1273 * has been zapped already via cleanup_highmem().
1275 all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1276 set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1278 #ifdef CONFIG_CPA_DEBUG
1279 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1280 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1282 printk(KERN_INFO "Testing CPA: again\n");
1283 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1286 free_kernel_image_pages((void *)text_end, (void *)rodata_start);
1287 free_kernel_image_pages((void *)rodata_end, (void *)_sdata);
1292 * Do this after all of the manipulation of the
1293 * kernel text page tables are complete.
1295 pti_clone_kernel_text();
1298 int kern_addr_valid(unsigned long addr)
1300 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1307 if (above != 0 && above != -1UL)
1310 pgd = pgd_offset_k(addr);
1314 p4d = p4d_offset(pgd, addr);
1318 pud = pud_offset(p4d, addr);
1322 if (pud_large(*pud))
1323 return pfn_valid(pud_pfn(*pud));
1325 pmd = pmd_offset(pud, addr);
1329 if (pmd_large(*pmd))
1330 return pfn_valid(pmd_pfn(*pmd));
1332 pte = pte_offset_kernel(pmd, addr);
1336 return pfn_valid(pte_pfn(*pte));
1340 * Block size is the minimum amount of memory which can be hotplugged or
1341 * hotremoved. It must be power of two and must be equal or larger than
1342 * MIN_MEMORY_BLOCK_SIZE.
1344 #define MAX_BLOCK_SIZE (2UL << 30)
1346 /* Amount of ram needed to start using large blocks */
1347 #define MEM_SIZE_FOR_LARGE_BLOCK (64UL << 30)
1349 /* Adjustable memory block size */
1350 static unsigned long set_memory_block_size;
1351 int __init set_memory_block_size_order(unsigned int order)
1353 unsigned long size = 1UL << order;
1355 if (size > MEM_SIZE_FOR_LARGE_BLOCK || size < MIN_MEMORY_BLOCK_SIZE)
1358 set_memory_block_size = size;
1362 static unsigned long probe_memory_block_size(void)
1364 unsigned long boot_mem_end = max_pfn << PAGE_SHIFT;
1367 /* If memory block size has been set, then use it */
1368 bz = set_memory_block_size;
1372 /* Use regular block if RAM is smaller than MEM_SIZE_FOR_LARGE_BLOCK */
1373 if (boot_mem_end < MEM_SIZE_FOR_LARGE_BLOCK) {
1374 bz = MIN_MEMORY_BLOCK_SIZE;
1378 /* Find the largest allowed block size that aligns to memory end */
1379 for (bz = MAX_BLOCK_SIZE; bz > MIN_MEMORY_BLOCK_SIZE; bz >>= 1) {
1380 if (IS_ALIGNED(boot_mem_end, bz))
1384 pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1389 static unsigned long memory_block_size_probed;
1390 unsigned long memory_block_size_bytes(void)
1392 if (!memory_block_size_probed)
1393 memory_block_size_probed = probe_memory_block_size();
1395 return memory_block_size_probed;
1398 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1400 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1402 static long __meminitdata addr_start, addr_end;
1403 static void __meminitdata *p_start, *p_end;
1404 static int __meminitdata node_start;
1406 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1407 unsigned long end, int node, struct vmem_altmap *altmap)
1416 for (addr = start; addr < end; addr = next) {
1417 next = pmd_addr_end(addr, end);
1419 pgd = vmemmap_pgd_populate(addr, node);
1423 p4d = vmemmap_p4d_populate(pgd, addr, node);
1427 pud = vmemmap_pud_populate(p4d, addr, node);
1431 pmd = pmd_offset(pud, addr);
1432 if (pmd_none(*pmd)) {
1436 p = altmap_alloc_block_buf(PMD_SIZE, altmap);
1438 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1442 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1444 set_pmd(pmd, __pmd(pte_val(entry)));
1446 /* check to see if we have contiguous blocks */
1447 if (p_end != p || node_start != node) {
1449 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1450 addr_start, addr_end-1, p_start, p_end-1, node_start);
1456 addr_end = addr + PMD_SIZE;
1457 p_end = p + PMD_SIZE;
1460 return -ENOMEM; /* no fallback */
1461 } else if (pmd_large(*pmd)) {
1462 vmemmap_verify((pte_t *)pmd, node, addr, next);
1465 if (vmemmap_populate_basepages(addr, next, node))
1471 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1472 struct vmem_altmap *altmap)
1476 if (boot_cpu_has(X86_FEATURE_PSE))
1477 err = vmemmap_populate_hugepages(start, end, node, altmap);
1479 pr_err_once("%s: no cpu support for altmap allocations\n",
1483 err = vmemmap_populate_basepages(start, end, node);
1485 sync_global_pgds(start, end - 1);
1489 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1490 void register_page_bootmem_memmap(unsigned long section_nr,
1491 struct page *start_page, unsigned long nr_pages)
1493 unsigned long addr = (unsigned long)start_page;
1494 unsigned long end = (unsigned long)(start_page + nr_pages);
1500 unsigned int nr_pmd_pages;
1503 for (; addr < end; addr = next) {
1506 pgd = pgd_offset_k(addr);
1507 if (pgd_none(*pgd)) {
1508 next = (addr + PAGE_SIZE) & PAGE_MASK;
1511 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1513 p4d = p4d_offset(pgd, addr);
1514 if (p4d_none(*p4d)) {
1515 next = (addr + PAGE_SIZE) & PAGE_MASK;
1518 get_page_bootmem(section_nr, p4d_page(*p4d), MIX_SECTION_INFO);
1520 pud = pud_offset(p4d, addr);
1521 if (pud_none(*pud)) {
1522 next = (addr + PAGE_SIZE) & PAGE_MASK;
1525 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1527 if (!boot_cpu_has(X86_FEATURE_PSE)) {
1528 next = (addr + PAGE_SIZE) & PAGE_MASK;
1529 pmd = pmd_offset(pud, addr);
1532 get_page_bootmem(section_nr, pmd_page(*pmd),
1535 pte = pte_offset_kernel(pmd, addr);
1538 get_page_bootmem(section_nr, pte_page(*pte),
1541 next = pmd_addr_end(addr, end);
1543 pmd = pmd_offset(pud, addr);
1547 nr_pmd_pages = 1 << get_order(PMD_SIZE);
1548 page = pmd_page(*pmd);
1549 while (nr_pmd_pages--)
1550 get_page_bootmem(section_nr, page++,
1557 void __meminit vmemmap_populate_print_last(void)
1560 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1561 addr_start, addr_end-1, p_start, p_end-1, node_start);